开启时会模糊预览图，关闭后正常显示
Of course, DYOR. Weight gain due to enhanced water retention, GI irritation, and headaches are the most commonly reported side effects. https://t.co/hp0Ult1gl5 https://t.co/GUJIBzE1EW https://t.co/JVcJt5WHll
But overall creatine is extremely well researched over long time frames, cheap, and readily available. If you take one supplement, make it creatine.
Why I take creatine and you should too 🧵 Most people think creatine is for gym bros. It's more than that. An energy and cognitive boost. Trying some new post types on consumer health as I build in the space. Goal is to share research so people live longer, healthier lives.
The strength effects of creatine are well studied. The chart above shows average differences in upper body lifts on creatine (4.4kg or ~10 pounds). I don't need to explain more here, everyone's familiar with this.
LLMs are not going to be AGI. We won't RL all the jobs out of existence with these things. The best thing you can say about LLMs => AGI is they will unlock the rest of whatever the actual AGI tech tree is and help us walk it.
Author. Coder. CTO. θηριομάχης. Building: https://t.co/otXT4Wy6WR. Writing: https://t.co/dBPBtyCIHw.
No. In ten years, people will still be doing many kinds of jobs and other people will be tweeting about how AGI is about to do all the jobs everywhere any year now.
LLMs are not going to be AGI. We won't RL all the jobs out of existence with these things. The best thing you can say about LLMs => AGI is they will unlock the rest of whatever the actual AGI tech tree is and help us walk it.
you tend to hear this a lot from people outside or new to ML, and I often point to a talk Ilya gave a few years back: 1) think of any decent deep neural net that has enough memory and sequential ops as just a big parallel computer 2) training this neural net is doing search over computer programs that maximize your objective 3)unless you have some large bottleneck (and given you can successfully optimize this system) you’ll find that these parallel computers are highly robust to architectural changes. 4) this is because computers are great at simulating each other. your new architecture can usually be straightforwardly simulated ‘inside’ your old architecture. 5) it’s not that architecture doesn’t matter, but it mostly matters with respect to (1) fundamental bottlenecks in this parallel computer (2) modifications that make models easier to optimize, since this argument only holds if your optimization is good (3) compute efficiency/system efficiency wins that make learning easier or faster. 6) it’s quite possible that new architectures will lead to breakthroughs in machine learning, but we should first start with bottlenecks, not naturalist intuitions about the ‘form’ of AI should take. until you understand this it seems surprising that small models trained longer are better than undertrained big models, that depth and width are surprisingly interchangeable, that talking to a model with an MoE or sparse attention or linear attention is approximately the same iso evals.
dei ex machina @openai, past: posttraining o3/4o, sora 1 & 2, applied research
RT @donpark: In my experience, DSPy is not just good for optimization but also great for unrooting pipeline issues much earlier, during the…
Asst professor @MIT EECS & CSAIL (@nlp_mit). Author of https://t.co/VgyLxl0oa1 and https://t.co/ZZaSzaRaZ7 (@DSPyOSS). Prev: CS PhD @StanfordNLP. Research @Databricks.